Borrowing Knowledge From Pre-trained Language Model: A New Data-efficient Visual Learning Paradigm
Wenxuan Ma, Shuang Li, Jinming Zhang, Chi Harold Liu, Jingxuan Kang, Yulin Wang, Gao Huang
摘要
The development of vision models for real-world applications is hindered by the challenge of annotated data scarcity, which has necessitated the adoption of dataefficient visual learning techniques such as semi-supervised learning. Unfortunately, the prevalent cross-entropy supervision is limited by its focus on category discrimination while disregarding the semantic connection between concepts, which ultimately results in the suboptimal exploitation of scarce labeled data. To address this issue, this paper presents a novel approach that seeks to leverage linguistic knowledge for data-efficient visual learning. The proposed approach, BorLan, Borrows knowledge from off-theshelf pretrained Language models that are already endowed with rich semantics extracted from large corpora, to compensate the semantic deficiency due to limited annotation in visual training. Specifically, we design a distribution alignment objective, which guides the vision model to learn both semantic-aware and domain-agnostic representations for the task through linguistic knowledge. One significant advantage of this paradigm is its flexibility in combining various visual and linguistic models. Extensive experiments on semi-supervised learning, single domain generalization and few-shot learning validate its effectiveness. Code is available at https://github.com/BIT-DA/BorLan .
问 Lune
几分钟读透这篇 Paper,不用再花一下午。
Lune 能读完全文,也能读引用它的顶会 Paper,讲清它的结论、证据和局限,每一点都注明出自哪篇 Paper。不用硬啃,也不靠猜。
已有 6 篇顶会 Paper 引用了它。别等审稿人问起,先弄清它们发现了什么。
脉络
- VeCAF: Vision-language Collaborative Active Finetuning with Training Objective Awareness 引用了 Swin Transformer: Hierarchical Vision Transformer using Shifted Windows
- VeCAF: Vision-language Collaborative Active Finetuning with Training Objective Awareness 引用了 Learning Transferable Visual Models From Natural Language Supervision
- UNIALIGN: Scaling Multimodal Alignment within One Unified Model 引用了 A ConvNet for the 2020s
- UNIALIGN: Scaling Multimodal Alignment within One Unified Model 引用了 Swin Transformer: Hierarchical Vision Transformer using Shifted Windows
- UNIALIGN: Scaling Multimodal Alignment within One Unified Model 引用了 Learning Transferable Visual Models From Natural Language Supervision
- A ConvNet for the 2020s 引用了 Swin Transformer: Hierarchical Vision Transformer using Shifted Windows
- Language-Assisted Debiasing and Smoothing for Foundation Model-Based Semi-Supervised Learning 引用了 Learning Transferable Visual Models From Natural Language Supervision
- Scaling Up Visual and Vision-Language Representation Learning With Noisy Text Supervision 引用了 A Simple Framework for Contrastive Learning of Visual Representations
- Scaling Up Visual and Vision-Language Representation Learning With Noisy Text Supervision 引用了 Learning Transferable Visual Models From Natural Language Supervision
- Swin Transformer: Hierarchical Vision Transformer using Shifted Windows 引用了 Learning Transferable Visual Models From Natural Language Supervision
- Learning Transferable Visual Models From Natural Language Supervision 引用了 A Simple Framework for Contrastive Learning of Visual Representations
- Learning Modality Knowledge Alignment for Cross-Modality Transfer 引用了 Swin Transformer: Hierarchical Vision Transformer using Shifted Windows
- Learning Modality Knowledge Alignment for Cross-Modality Transfer 引用了 Learning Transferable Visual Models From Natural Language Supervision
引用它的顶会 Paper6
- Learning Modality Knowledge Alignment for Cross-Modality TransferWenxuan Ma, Shuang Li, Lincan Cai, Jingxuan KangICML 2024 · 10 次引用
- VeCAF: Vision-language Collaborative Active Finetuning with Training Objective AwarenessRongyu Zhang, Zefan Cai, Huanrui Yang, Zidong Liu 等ACM MM 2024 · 5 次引用
- PLMTrajRec: A Scalable and Generalizable Trajectory Recovery Method with Pre-trained Language ModelsTonglong Wei, Yan Lin, Youfang Lin, Shengnan Guo 等NeurIPS 2025 · 3 次引用
- UNIALIGN: Scaling Multimodal Alignment within One Unified ModelBo Zhou, Liulei Li, Yujia Wang, Huafeng Liu 等CVPR 2025
- Language-Assisted Debiasing and Smoothing for Foundation Model-Based Semi-Supervised LearningNa Zheng, Xuemeng Song, Xue Dong, Aashish Nikhil Ghosh 等CVPR 2025
它引用的 Paper30
- Learning Transferable Visual Models From Natural Language SupervisionAlec Radford, Jong Wook Kim, Chris Hallacy, Aditya Ramesh 等ICML 2021 · 47,906 次引用
- Swin Transformer: Hierarchical Vision Transformer using Shifted WindowsZe Liu, Yutong Lin, Yue Cao, Han Hu 等ICCV 2021 · 31,683 次引用
- A Simple Framework for Contrastive Learning of Visual RepresentationsTing Chen, Simon Kornblith, Mohammad Norouzi, Geoffrey E. HintonICML 2020 · 24,064 次引用
- A ConvNet for the 2020sZhuang Liu, Hanzi Mao, Chao-Yuan Wu, Christoph Feichtenhofer 等CVPR 2022 · 6,782 次引用
- Scaling Up Visual and Vision-Language Representation Learning With Noisy Text SupervisionChao Jia, Yinfei Yang, Ye Xia, Yi-Ting Chen 等ICML 2021 · 5,401 次引用
相关 Paper
- Fine-Grained Visual Prompt Learning of Vision-Language Models for Image RecognitionHongbo Sun, Xiangteng He, Jiahuan Zhou, Yuxin PengACM MM 2023 · 16 次引用
- Training Spatial-Frequency Visual Prompts and Probabilistic Clusters for Accurate Black-Box Transfer LearningWonwoo Cho, Kangyeol Kim, Saemee Choi, Jaegul ChooACM MM 2024
- Data-Efficient Language Shaped Few-shot Image ClassificationZhenwen Liang, Xiangliang ZhangEMNLP Findings 2021
- Language Does Matter for Cross-Domain Few-Shot Visual Feature EnhancementFei Zhou, Xiwen Zhang, Qingqing Qiu, Lei Zhang 等CVPR 2026
- Semi-Supervised CLIP Adaptation by Enforcing Semantic and Trapezoidal ConsistencyKai Gan, Bo Ye, Min-Ling Zhang, Tong WeiICLR 2025
